IARC 50th Anniversary Conference
Causality and chance in the origins of cancer
Dr George Davey Smith - University of Bristol, Bristol, UK
There’s been a lot of attention paid to the idea that cancer is bad luck, in particular when a very high profile paper came out at the beginning of last year which said that because mutations are occurring during cell divisions that they appear to underlie cancers and also indicated which sort of tissues, which organs of the body, we get the most cancers in. Then one of the popular messages that was taken from that was that the majority of cancer is just bad luck. In fact when the news story was on the radio, British radio, the BBC Radio 4 Today programme, some person tweeted a photograph of themselves smoking and they said, ‘Thank you BBC Radio 4 for this story.’
I think there’s a serious misunderstanding about the causes of cancer at an individual level if you’re actually trying to say why did this individual person develop cancer than if you were going to say at a population, if you say why do some groups of people have much higher risks of cancer than other groups of people. The actual causes for individual level may be very different from a group level cause so, of course, if everyone smokes. So a thought experiment, as I come from Bristol, is if you imagine that in Bristol a law was passed that when everyone got up in the morning they were given a pack of twenty cigarettes and they were not allowed to go to bed until they’d smoked all twenty cigarettes and then in Bath, which is a town nearby, cigarettes were completely banned, no-one smoked cigarettes at all. Then you follow the towns up for thirty years and then thirty years later there would be a much higher rate of lung cancer in Bristol than in Bath, a much higher rate. All of which is clearly attributable to smoking, that’s the only thing which is now different, has been made different between the two places. But in Bristol everyone smokes twenty cigarettes a day so at the individual level you couldn’t show that smoking is related to lung cancer. It can’t be, everyone’s smoking exactly twenty cigarettes a day. So for those individuals there will be elements of luck, possibly genetic contributions etc., those will be the things determining who in Bristol gets cancer would be luck and genes.
Precisely because of the fact that at the individual level luck and genes etc. influence who does get cancer, who is subjected to the same sort of exposures, that’s why you get the sort of Uncle Norman idea, you know, ‘My Uncle Norman smoked all his life and he lived to 95. How can smoking be bad for you?’ But that’s not actually contradictory to the fact that at a population level virtually all of the cases of lung cancer could be attributed to smoking. So I don’t think it’s any good going out and telling people they’re just wrong, thinking that chance has a role, it’s a complicated message but you need actually to get the complicated message over which is that at the population level you could reduce cancer risks dramatically by changing some of these exposure patterns.
Given that your interest is in genes and this “chance process”, what is your message of hope?
I think the hope is the fact that you can demonstrate, at a group level you can demonstrate the causes of a large proportion of cancers. The way that genes could contribute to this is that it’s very difficult studying a lot of behavioural risk factors. As you know, one day you see a story in the newspapers saying that red wine is good for you and the next day that it’s bad for you and such like. Because behaviours like diet and like many other behavioural factors they relate very strongly to people’s social circumstances, for example, which influence their risk. They are interrelated, people who have a bad diet might be more likely to smoke, might be more likely to drink heavily, might be more likely to have acquired infections that relate to cancer, etc. So it’s very difficult separating out things which are causes from things which are just markers of those causes. In epidemiology we call this confounding and there has been dramatic demonstrations of that in the chronic disease world over the last twenty years, twenty or thirty years, which is when anti-oxidant vitamins, for example, which a lot of observational studies suggested that higher levels of beta carotene were good for you, higher levels of some forms of vitamin A were good for you etc., higher levels of vitamin C. But then when randomised controlled trials were done, randomising people to a different level of intake of these vitamins they had absolutely no effect on the outcomes.
What are the practical messages that can be used in the midst of all this variability?
The way that genes can help us separate out the causal factors from the non-causal factors is that if you have genetic variants that relate to these exposures, like smoking, like your anti-oxidant vitamin levels, like your alcohol drinking, and we have those variants, we have genetic variants related robustly to all of those things, then if those factors are causal then the genetic variant which relates, for example, to more smoking will be related to a higher level of any disease which smoking causes. This has been very strongly demonstrated, that genetic variants which influence people such that at a group level, on average, people who carry one variant rather than the other smoke more, and they smoke more heavily, the people who carry that gene have a higher risk of lung cancer to the extent that you’d expect because of their smoking behaviour.
Are there any further examples?
Genetic variants related to some anti-oxidant vitamins have not related to lower risk of some outcomes. Genetic variants related to alcohol have been used extensively in this form of study which is called Medelian randomisation. They have demonstrated that, contrary to popular opinion, there’s no good evidence that alcohol is protective against cardiovascular disease; alcohol definitely increases blood pressure and alcohol increases the risks of oesophageal cancer and some other forms of head and neck cancer. So these things which are difficult to study observationally because people who drink more might have better or less good health behaviours in other domains so it’s very difficult to separate those things, using the genetic variants in this way gives you really strong evidence or stronger evidence of whether they’re causes or not.
What is your take-home message?
Prevention at a population level which is based upon reliable evidence of what factors really do influence risk has huge potential. We need to apply methods which separate true causal factors from all these other confounding factors to be able to give reliable messages and to be able to establish programmes which really would be effective.